Skip to content
Blog

What is SLAM? Mapping and localization for drones

Published: · Updated:

SLAM (simultaneous localization and mapping) is how a robot or drone builds a map of an environment it has never seen while estimating its own position in that map at the same time. It is one of the building blocks of navigation where GPS is unavailable or unreliable.

What problem does SLAM solve?

To know where a vehicle is, you need a map; to build an accurate map, you need to know where the vehicle was when it took each measurement. SLAM solves the two questions together rather than one after the other: every new sensor measurement updates both the map and the vehicle's position and orientation within it.

The result is a position estimate computed entirely on board, without satellite signals or a map prepared in advance. That position is usually relative to a local coordinate frame; global latitude and longitude require an additional reference.

How does SLAM work?

Most SLAM systems have two layers. The front end processes raw sensor data: it extracts distinctive points from camera images or aligns LiDAR point clouds, and it matches the same point across different measurements (data association).

The back end uses those matches to estimate the vehicle's path and the map together. Early systems used filter-based methods such as the extended Kalman filter (EKF-SLAM). Today, optimization-based methods such as pose graphs and bundle adjustment are common.

The step that most sets SLAM apart from odometry is loop closure. When the vehicle recognizes a place it has visited before, it can correct the error accumulated along the way in one go. This keeps long-term drift in check.

Which sensors are used?

SLAM is not tied to a particular sensor; there are different kinds depending on the sensor used:

  • Visual SLAM uses a single camera (monocular), a stereo camera or a depth camera (RGB-D). It is light and low-power, but sensitive to lighting and ground texture.
  • LiDAR SLAM measures the geometry of the surroundings directly with a laser scanner. It does not depend on light, but the sensor is heavier and more expensive.
  • Visual-inertial SLAM combines a camera with an inertial measurement unit (IMU). The IMU handles fast motion and helps resolve the scale ambiguity of single-camera systems.

Challenges of SLAM on drones

Many SLAM methods developed for ground robots run into new problems when moved to aircraft:

  • Computing power and energy: the computation has to run in real time on a small, light, low-power computer.
  • Speed and motion blur: in fast flight, consecutive frames differ more and images can blur.
  • Textureless ground: surfaces such as water, snow, sand or flat fields offer few distinctive points to track.
  • Light and weather: night, sunset, fog and shadows make visual methods harder.
  • Altitude: the higher the aircraft, the smaller the ground details, and the harder depth and scale are to estimate.
  • Chances for loop closure: on long-range flights the aircraft rarely returns to the same place, so there may be few moments to correct accumulated error.

SLAM, VIO and other GPS-denied methods

Visual-inertial odometry (VIO) also uses a camera and an IMU, but it usually does not keep a persistent map or perform loop closure. That makes it lighter, but its error grows over time. SLAM can correct that error when the vehicle returns to a known area.

Because loop closure is not always possible on long-range, one-way flights, practical systems often combine SLAM or VIO with sources such as terrain-referenced navigation (TRN), scene matching and inertial navigation through sensor fusion.

An overview of GPS-denied navigation methods →

Frequently asked questions

GPS provides a global, absolute position from satellite signals, so it depends on a signal from outside. SLAM estimates position with the vehicle's own sensors, relative to a local map the vehicle builds itself, and needs no outside signal.

Yes, by definition SLAM does not need GPS. The position it produces is relative to a local coordinate frame, though; tying it to global coordinates requires a reference such as a known starting point, map matching or GPS when it is available.

A single camera reduces the scene to a two-dimensional image and does not measure distance directly. A camera-only system can therefore find the direction of motion but not the true distance. Scale is resolved with extra information such as an IMU, a stereo camera, an altimeter or an object of known size.

Loop closure is when the vehicle recognizes a place it has passed before and uses that to correct the position error accumulated along the way. It is the key feature that separates SLAM from methods that only do odometry.